Ricardo T. Lemos, Bruno Sansó and Marc Los Huertos
12/31/2006 09:00 AM
Applied Mathematics & Statistics
We consider monthly temperature data collected over a period of sixteen years at 24 locations in the Elkhorn Slough National Estuary, located in the Monterey Bay area in Central California, USA. Our goal is to develop a statistical model in order to examine spatial and temporal variation in water quality in the slough. We focus on the detection of trends. Exploratory data analysis reveals the presence of warming trends which vary according to the different locations. We develop a model that is based on mixing two seasonal temperature components; one serves as a proxy for coastal behavior and the second serves as a proxy for inland behavior. Each location corresponds to a mixture with time varying weights. Additionally, location dependent trends are considered, together with a common time varying baseline. The variability of such a baseline is linearly dependent on a variable that summarizes several atmospheric measurements. We use a Bayesian approach with a purposely developed Markov chain Monte Carlo method to explore the posterior distribution of the parameters. We find that seasonal patterns have changed in time, that neighboring stations can have substantially different behaviors and that most stations show significant warming trends.
